Onit SideKick – AI Chat Sidebar for Any Mac App

TL;DR

TL;DR: Onit SideKick is a macOS app that attaches an AI chat sidebar to any application, letting you use Claude, GPT-4.5, Gemini, or local Ollama models directly from whatever app you are working in.

Source and Accuracy Notes

⚠️ This section is MANDATORY. All links must be verified from actual source, not guessed.

• Project page: getonit.ai/ai-sidebar ← verified
• Source repository: github.com/synth-inc/onit ← verified README
• License: NOASSERTION (source-available, not OSI-approved) ← verified via GitHub API
• HN launch thread: news.ycombinator.com/item?id= ← original HN post not found in Algolia; using HN announcement only
• Source last checked: 2026-07-09 (commit verified against README)

What Is Onit SideKick?

Onit SideKick is a macOS application that installs as a persistent sidebar you can dock to any window. Unlike Cursor Chat which lives inside an IDE, SideKick works alongside any app — your browser, a terminal, a code editor, a PDF reader, or anything else.

The core interaction is straightforward: select text anywhere, press a hotkey (default Command+0), and the sidebar opens with your selected text already loaded as context. You can also enable Auto-Context, which automatically reads the content of the docked window as context before you hit submit.

Under the hood it is provider-agnostic. You bring your own API keys for remote providers, and it also ships with a Local Mode that connects directly to Ollama so your conversations never leave your machine.

Setup Workflow

Step 1: Install Ollama (optional, for local mode)

If you want to run models locally, install Ollama first:

brew install ollama

After installation, pull a model:

ollama pull llama3.2

Ollama starts automatically as a background service and exposes a local API that Onit SideKick auto-detects.

Step 2: Install Onit SideKick

Download the pre-built macOS app from the project page:

https://www.getonit.ai/ai-sidebar

Open the downloaded .app and grant the required permissions (Accessibility for hotkey interception, Screen Recording if you use Auto-Context window-capture).

Step 3: Configure API Providers

1. Launch Onit SideKick
2. Open Settings (gear icon)
3. Under Providers, add your API keys for the services you want to use:
   • Anthropic (Claude)
   • OpenAI (GPT-4.5, o3)
   • GoogleAI (Gemini 2.5)
   • Perplexity (Sonar, Sonar Deep Research)
   • Deepseek (R1)
   • xAI (Grok)
   • Custom — OpenRouter, Groq, and any OpenAI-compatible endpoint
4. For local mode, ensure Ollama is running and toggle Local Mode on

Step 4: Set Your Hotkey

Default hotkey is Command+0. You can change it in Settings under Shortcuts. Choose any unused macOS shortcut combination.

Step 5: Start Using It

• Highlight text in any app → press your hotkey → the sidebar opens with your selection pre-loaded
• Dock a window to the sidebar → toggle Auto-Context → the window content is captured as context
• Swap models mid-conversation via the model picker dropdown inside the sidebar

Practical Evaluation Checklist

• [ ] Sidebar attaches to non-developer apps (Safari, Notes, Mail)
• [ ] Highlight-to-chat workflow works across app boundaries
• [ ] Auto-Context captures window content without manual copy-paste
• [ ] Local Mode via Ollama routes traffic to localhost only
• [ ] Multiple remote providers work in the same session (provider switching)
• [ ] Hotkey does not conflict with other macOS shortcuts
• [ ] API keys are stored locally and not sent to Onit’s servers when self-provided

Security Notes

• Local Mode: When enabled, all inference traffic stays on your machine via Ollama’s local API. No data leaves localhost.
• Remote Mode (self-key): API calls go directly to Anthropic/OpenAI/Google/etc. with your own credentials. Onit’s servers are not in the request path.
• Remote Mode (no key): If you do not add your own API token, requests are proxied through Onit’s server. They state they do not store request contents.
• Auto-Context: Window content is captured locally and only uploaded when you press Submit. It is not continuously streamed.
• Crash analytics: PostHog and Firebase Crashlytics are enabled by default; both can be disabled in Settings.

FAQ

Q: Is Onit SideKick open source? A: The repository is source-available on GitHub under a NOASSERTION license. The projects for Onit Dictate and Onit QuickEdit are in separate private repositories and are not open source.

Q: Can I use it without an API key? A: Yes, but functionality is limited. Without your own API key for remote providers, requests are routed through Onit’s proxy server. For full privacy and unlimited usage, use Local Mode with Ollama.

Q: Which local models does it support? A: Any model available in your local Ollama installation. Common choices include Llama 3.2, Mistral, and Phi. Onit SideKick detects your Ollama library automatically.

Q: Does it work on Intel Macs? A: The product page targets Apple Silicon. Performance on Intel Macs has not been verified — Ollama itself supports Intel but the Onit SideKick app binary may require ARM.

Q: Is Auto-Context screen recording? A: It captures the content of the docked window via macOS accessibility APIs, not a full screen recording. It only reads the window content when you explicitly trigger a submit, not continuously.

Conclusion

Onit SideKick solves the “I want AI help right here in this app” problem that Cursor, Copilot, and other IDE-bound tools leave unsolved. The provider-agnostic design means you are not locked into one model family, and Local Mode via Ollama makes it viable for privacy-sensitive workflows without杰 switching tools.

If you spend your day jumping between a code editor, a browser, a terminal, and a PDF reader, docking an AI sidebar to each of those windows is a meaningfully different interaction model than constantly context-switching to a chat app. It is worth trying if you already use Claude or GPT for multi-app workflows.

Download: getonit.ai/ai-sidebar